DOI: 10.3303/CET25120086 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
Paper Received: 14/04/2025; Revised: 25 August 2025; Accepted: 28 September 2025 
Please cite this article as: Bian R., He R., Yan X., Tong X., 2025, LSTM-Driven Predictive Scheduling for Green Energy-Hydrogen-Methanol 
Integrated System, Chemical Engineering Transactions, 120, 511-516  DOI:10.3303/CET25120086 
  

 CHEMICAL ENGINEERING TRANSACTIONS  
 

VOL. 120, 2025 

A publication of 

 

The Italian Association 
of Chemical Engineering 
Online at www.cetjournal.it 

Guest Editors: Bing Shen How, Viknesh Andiappan, Denny K.S. Ng, Hon Loong Lam, Petar S. Varbanov 

Copyright © 2025, AIDIC Servizi S.r.l. 

ISBN 979-12-81206-21-2; ISSN 2283-9216 

LSTM-Driven Predictive Scheduling for Green Energy-

Hydrogen-Methanol Integrated System 

Rui Biana, Renchu Hea,*, Xinyu Yana, Xinglin Tongb 

aDepartment of Automation, College of Artificial Intelligence, China University of Petroleum, Beijing, 102249, China  
bWuhan University of Technology, Wuhan, 430070, China  

 rche@cup.edu.cn  

This study proposes an optimization scheduling model for the Green Energy-Hydrogen-Methanol Integrated 

System (GEHMIS) based on the Long Short-Term Memory (LSTM) network, aimed at achieving efficient, stable, 

and economical operation of renewable energy systems. In response to the fluctuations and uncertainties in the 

output of wind and solar energy, this paper introduces the LSTM network to accurately predict wind speed and 

temperature using historical data. The integrated system model developed in this study covers key processes 

such as wind and solar power generation, water electrolysis for hydrogen production, hydrogen storage and 

transportation, and methanol synthesis. The objective of this research is to maximize the overall system profit 

by developing a multi-constraint mathematical optimization model that takes into account multiple factors, 

including hydrogen and methanol production efficiencies, hydrogen consumption, operation and maintenance 

costs, electricity price fluctuations, grid interactions, and carbon dioxide costs. Compared to traditional random 

resource allocation strategies, the proposed optimization model demonstrates significant advantages in terms 

of economic benefits and energy utilization efficiency: the overall system profit increased by approximately 68.3 

%, and the hydrogen utilization rate improved by 30 %. This provides important technical support and a practical 

pathway for the large-scale application of green energy and the global transition to sustainable energy. 

1. Introduction 

Amid growing global energy crisis and environmental challenges, the over-reliance on traditional energy sources 

has created pressing issues, necessitating urgent development of clean and sustainable energy solutions 

(Wang et al., 2023). While renewable energies like wind and solar power face integration and storage challenges 

in large-scale applications, hydrogen and methanol have emerged as promising clean energy carriers (Mazzeo 

et al., 2022). The wind-solar hydrogen-to-methanol technology not only enhances renewable energy utilization 

but also contributes to CO₂ reduction, offering significant environmental advantages. 

However, the intermittent nature of wind and solar power complicates accurate forecasting and scheduling using 

conventional methods. Long Short-Term Memory (LSTM) networks, a specialized Recurrent Neural Network 

(RNN) architecture introduced by Hochreiter and Schmidhuber (1997) in 1997, demonstrated exceptional 

capability in capturing long-term dependencies and have been widely adopted for renewable energy forecasting 

(Sherstinsky, 2020). Al-qaness et al. (2024) achieved improved wind power prediction accuracy by integrating 

LSTM with optimization algorithms. In the context of energy economic efficiency, the concept of the methanol 

economy proposed by Prakash et al. (2011) regarded the hydrogenation of CO₂ to methanol as one of the most 

attractive and potentially profitable technological pathways. To realize the systematic application of this pathway, 

Macedo and Peyerl (2022) conducted an economic analysis of hydrogen production from wind-solar power 

plants in Brazil, demonstrating that selling hydrogen is more economically beneficial than converting it into 

electricity. Zheng et al. (2022) constructed a model of the production system and quantitatively evaluated the 

influence of key operational parameters on methanol cost structures. 

Currently, most wind and solar power generation methods still rely on static or low-accuracy weather forecasting 

models, which struggle to capture the dynamic nature of renewable energy. Although some studies have 

explored green electricity-based hydrogen production pathways, few have addressed specific strategies for 

equipment scheduling and hydrogen processing. Moreover, existing research mainly focuses on hydrogen itself, 

511



lacking comprehensive analysis of co-production with downstream products, such as methanol, and related 

market revenues, resulting in limited model practicality and incomplete assessment of system-level economic 

performance. 

To address these issues, this paper proposes an LSTM-based optimization method for scheduling in GEHMIS. 

The LSTM network, trained on historical data, is used to predict wind speed and temperature, enabling dynamic, 

multi-constraint optimization of hydrogen production and methanol synthesis. The model aims to maximize 

overall profit by optimizing power allocation and ensuring stable and efficient system operation. This approach 

offers the potential for cost reduction, emission mitigation, and improved economic performance, providing both 

theoretical and practical value. 

2. Modeling methodology for the GEHMIS 

The GEHMIS primarily comprises the wind-solar power generation system, hydrogen production by electrolyzer, 

grid interaction, and methanol production and sales. Hydrogen is partly sold and partly used with CO₂ to produce 

methanol. Renewable power is prioritized for electrolyzer, with surplus sold; when insufficient, grid electricity is 

purchased to maintain stable operation. The specific schematic diagram is shown in Figure 1. 

Photovoltaic Power 
Generation

Wind Power 
Generation

Electrolyzer

Methanol Production Methanol Sales

Power Grid

CO2

Electric Power 
Flow

Hydrogen 
Flow

CO2 
Flow

Methanol 
Flow

Hydrogen Sales

 

Figure 1: The structural diagram of the GEHMIS 

2.1 System mathematical models 

The study establishes a a wind–solar–hydrogen–methanol model and enables interaction with the power grid. 

(1) Wind power model  

1 3

2 3

, 1 2

0 2 3

0

1

2
WD t P

v v or v v

P R C v v v v

P v v v



   



=  

  

  (1) 

Eq(1) represents the wind power model based on the works of Wang et al. (2024). The output power of the wind 

turbine 
,WD tP  over a time period t is jointly determined by the radius of the turbine blades R, wind energy 

utilization coefficient PC , wind speed v, and the rated power 0P . 1 2 3, ,v v v   represent the cut-in wind speed, 

rated wind speed and cut-out wind speed of the wind turbine, respectively. 

(2) Photovoltaic power model 

=  + −  , [1 ( 25)]PV t STC t PV PVP P k T S   (2) 

The photovoltaic power model in Eq(2) is derived from the methodology proposed by Yousefi et al. (2017), 

where, 
,PV tP  denotes the photovoltaic power output and tT  represents surface temperature. STCP  represents 

the light intensity coefficient, k is the power temperature coefficient, PV  is the conversion efficiency of the 

photovoltaic array, and PVS  is the area of the photovoltaic array. 

(3) Electrolyzer and methanol synthesis section models  

Eq(3) and Eq(4) formulate the electrolyzer and methanol synthesis section models based on the works of Buttler 

and Spliethoff (2018). 

512



 =    =  2 4 4 4

2 4
,H CH CH CHE PEM

t t H t t CHq P L q P L   (3) 

 =   =4 4 2 4 2, /CH CH CO CH COCHH

t t t tq q q V   (4) 

where,  2 4, HH

t

C

t
qq  represent the hydrogen and methanol production,   4, CHE  represent the production efficiency 

of hydrogen and methanol, 4, CHPEM

t tP P  represent the power consumption of the electrolyzer and the methanol 

production equipment, and  
2 4
,H CHL L  represent the electricity-to-hydrogen and electricity-to-methanol factor. 

CHH

tq  refers to the hydrogen required for methanol production, while  4CH
 refers to the hydrogen consumption 

per kg of methanol produced. 2CO

tq  is the required production of CO2, and  2CO  refers to the carbon dioxide 

utilization rate. 

2.2 System optimization models 

This study focuses on short-term scheduling issues. Since the initial investment cost is a fixed expenditure, it is 

not considered. This study also disregards system start-up/shutdown costs and separate transportation costs. 

2
max revenu inve M COC C C C C= − − −  (5) 

where, Eq(5) is the objective function of the model. C represents the total revenue, revenueC  includes the sales 

of hydrogen and methanol, while invC  represents the operating costs, MC  is the grid interaction costs, and 
2COC  

represents the cost of CO2. 

(1) Revenue calculation 

=

=  +  2 _2 4 4

t t
1

( )sell

revenue

T
HH CH CH

t

C c q c V  (6) 

where, 2Hc  and 4CHc  represent the market selling prices of hydrogen and methanol, and 2 _H sell

t
q  represents the 

quantity sold of hydrogen. 

(2) Cost calculation 


= =

=  
4

1 1

( ( ))
T

t

inv i i

t i

C c  (7) 

 
=

=  − 
1

( )
T

buy sell

M buy t sell t

t

C P P  (8) 


=

=  2

2 2

1

)( CO

t

T

CO CO

t

qC  (9) 

where, i  represents the various system equipment, including wind power generation, photovoltaic generation, 

electrolyzer, and methanol production units. ic  represent the operating costs of each type of equipment, and 

 t

i
 represent the consumption of each equipment. ,buy sell   and 

2CO  represent purchasing, selling price of the 

electricity, and carbon dioxide purchasing price. ,buy sell

t tP P  represent the power associated with buying and 

selling electricity, respectively. 

The model constraint equations are as follows: 

(1) Power constraint 

+ + = + + 4

, ,

CHbuy sell PEM

WD t PV t t t t tP P P P P P  (10) 

The total of generated and purchased electricity equals the sum of electricity consumption and electricity sold. 

(2) Hydrogen balance constraint 

+ =2 2 4 2_ ,H sell H CH H

t t tq q q  (11) 

513



The hydrogen produced by the electrolyzer 2H

tq  is supplied for methanol production 2 4,H CH

tq  and sold for 

hydrogen use 2 _H sell

tq . 

(3) Interaction and power constraints 

min maxPEM

PEM t PEMP P P    (12) 

 4

4 4

min maxCH

CH t CHP P P  (13) 





   


  − 

min max

,

min max

,(1 )

buy

M t M t M

sell

M t M t M

P P P

P P P
 (14) 

where,      
4 4

min max min max min max, , , , ,PEM PEM CH CH M MP P P P P P  are the minimum and maximum power of the electrolyzer, methanol 

production equipment, and grid interaction.  ,M t  represents the state of purchasing and selling electricity from 

the superior power grid, where a value of 1 indicates purchase mode and 0 indicates sale mode. 

3. Case study and system parameters 

This case study selects Cangzhou, Hebei Province, China, as the target region. Based on wind speed and 

temperature data predicted by the LSTM neural network model, a dynamic optimization scheduling analysis of 

the GEHMIS was conducted over a two-month period, using a one-day time step. The key components involved 

in the system include wind turbine, photovoltaic module, electrolyzer, and methanol synthesis unit, with their 

core unit costs in Table 1 (Wiser et al., 2019). In terms of economic parameters, the selling price of methanol is 

set at 2.6 CNY/kg, the cost of CO₂ capture is 0.40 CNY/kg, and the hydrogen selling price is 20 CNY/kg (Liu et 

al., 2023). Meanwhile, the system is connected to the main power grid, with average electricity purchase price 

of 0.3 CNY/kWh and selling price of 0.15 CNY/kWh (Yang et al., 2023). 

Table 1: Costs of each equipment 

Equipment Operating cost 

Wind turbine 0.04 CNY/kWh 

Photovoltaic 0.05 CNY/kWh 

Electrolytic cell 0.2 CNY/kWh 

Methanol production equipment 0.528 CNY/kg 

4. Analysis of system results 

This section primarily focuses on conducting an in-depth analysis of the LSTM prediction results and the system 

optimization scheduling results obtained by using the CPLEX optimizer based on the parameter settings. 

4.1 LSTM-based prediction results 

To evaluate the forecasting accuracy of the LSTM model, the prediction performance and error trends are 

analyzed to evaluate the method’s reliability. The results are shown in Figure 2. Figure 2(a) presents a 

comparison between the actual and LSTM-predicted wind speed values, with the prediction errors ranging from 

-0.036 to 0.020. Figure 2(b) depicts temperature prediction results, where the errors between predicted and 

actual values range from -0.094 to 0.131. 

a b
 

Figure 2: Prediction results of (a) LSTM wind speed, (b) temperature 

514



These error values denote the point-wise deviation between predicted and actual values, expressed in the same 

units as the target variables. The mean absolute percentage error (MAPE) was introduced to systematically 

evaluate overall model performance, where lower values indicate higher predictive accuracy. It is defined as: 

=

−
= 

1

ˆ1
| | 100%

n
i i

i i

y y
MAPE

n y
 (15) 

where, 
iy  denotes the actual observed value, ˆ

iy  represents the predicted value, and n  is the total number of 

samples. Both MAPE value of wind speed and temperature predictions are below 0.3 %, demonstrating the 

model's exceptionally strong forecasting accuracy. 

4.2 Optimized scheduling results of the GEHMIS 

The optimized system operation is analyzed, including equipment scheduling and power transactions, in 

accordance with energy and hydrogen balance principles. Figure 3(a) illustrates the overall power balance 

scheduling, where wind and photovoltaic systems generate a total of 1.423×105 kWh during the cycle. Figure 

3(b) presents the hydrogen scheduling balance, where the electrolyzer produces 5.877×104 kg of hydrogen, of 

which 3.996×104 kg is sold. The remaining hydrogen is used for methanol synthesis, yielding 1.8076×104 kg of 

methanol. The overall hydrogen utilization rate is approximately 32 %. 

a b  

Figure 3: (a) Electrical power, (b) hydrogen balance scheduling diagram of the optimized model 

a b  

Figure 4: (a) Electrical power, (b) hydrogen balance scheduling diagram of the stochastic model 

ba
 

Figure 5: The revenue and cost of the (a) optimized model, (b) stochastic model 

According to Figure 5(a), the total system revenue amounts to 9.86×105 CNY. Among this, hydrogen sales are 

the primary source of income, generating 7.995×105 CNY, which accounts for approximately 81 % of the total 

revenue. The final system profit is 1.237×105 CNY. To further validate the advantages of the optimized 

scheduling model, a comparative analysis was conducted against a stochastic model under the same cost 

parameters. As shown in Figure 4(b), the electrolyzer produces a total of 9.936×104 kg of hydrogen, of which 

515



only 2 % is used for methanol synthesis. Figure 5(b) shows that the stochastic model yields a final profit of 

7.348×104 CNY, while the optimized model achieves a 68.3 % increase in profit. In addition, the hydrogen 

utilization rate rises from 2 % to 32 %, indicating a higher share of hydrogen being used for methanol production. 

These results demonstrate that, under all operational constraints, the optimized scheduling model significantly 

enhances the economic performance and energy conversion efficiency. 

5. Conclusions 

Experimental research on the GEHMIS based on LSTM prediction demonstrates excellent performance in wind 

speed and temperature prediction, with MAPE below 0.3 %. Compared to the random allocation model, the 

optimized scheduling model boosts economic benefits by 68.3 %, increases hydrogen utilization rate by 30 %. 

These results demonstrate that the optimized model significantly improves system operational efficiency and 

economic performance, providing strong support for stable operation and efficient energy conversion. Future 

research should focus on integrating multi-objective optimization algorithms and improving real-time scheduling 

to enhance system performance. These advances will support large-scale green energy deployment and aid 

the global energy transition toward sustainable development goals. 

Acknowledgments 

This work was supported by National Natural Science Foundation of China (62073142), Science Foundation of 

China University of Petroleum, Beijing (No.2462024YJRC011) and the Open Research Project of the State Key 

Laboratory of Industrial Control Technology, China (Grant No. ICT2024B70). Xinglin Tong is an adjunct 

professor at the Department of Automation, College of Artificial Intelligence, China University of Petroleum, 

Beijing. 

References 

Al-qaness M.A.A., Ewees A.A., Aseeri A.O., Elaziz M.A., 2024, Wind power forecasting using optimized LSTM 

by attraction-repulsion optimization algorithm. Ain Shams Engineering Journal, 15, 103150. 

Buttler A., Spliethoff H., 2018, Current status of water electrolysis for energy storage, grid balancing and sector 

coupling via power-to-gas and power-to-liquids: A review. Renewable and Sustainable Energy Reviews, 82, 

2440–2454. 

Hochreiter S., Schmidhuber J., 1997, Long short-term memory. Neural computation, 9, 1735–1780. 

Liu K., Zhou J., Xu G., Chen Y., Zhang R., Liu W., 2023, Planning and Scheduling of Integrated System of Light 

Hydrogen Storage and Methanol for High Proportion of Renewable Energy Consumption. Journal of Chinese 

Society of Power Engineering, 43, 1477–1486. (in Chinese) 

Mazzeo D., Herdem M.S., Matera N., Wen J.Z., 2022, Green hydrogen production: Analysis for different single 

or combined large-scale photovoltaic and wind renewable systems. Renewable Energy, 200, 360–378. 

Macedo S. F., Peyerl D.,2022, Prospects and economic feasibility analysis of wind and solar photovoltaic hybrid 

systems for hydrogen production and storage: A case study of the Brazilian electric power sector. 

International Journal of Hydrogen Energy, 2022, 47, 10460–10473. 

Prakash G.K.S., Olah G., Goeppert A., 2011, Beyond oil and gas: The methanol economy. ECS Transactions, 

35, 31. 

Sherstinsky A., 2020, Fundamentals of recurrent neural network (RNN) and long short-term memory (LSTM) 

network. Physica D: Nonlinear Phenomena, 404, 132306. 

Wang J., Zhang C., Ai S., Yang X., Ge M., 2024, Modeling and analysis of multi-energy isolated microgrid with 

wind-PV-diesel-storage in islands and reefs. Journal of Naval University of Engineering, 36, 79–86. (in 

Chinese) 

Wang Y., Song M., Jia M., Li B., Fei H., Zhang Y., Wang X., 2023, Multi-objective distributionally robust 

optimization for hydrogen-involved total renewable energy CCHP planning under source-load uncertainties. 

Applied Energy, 342, 121212. 

Wiser R., Bolinger M., Lantz E., 2019, Assessing wind power operating costs in the United States: Results from 

a survey of wind industry experts. Renewable Energy Focus, 30, 46–57. 

Yousefi H., Ghodusinejad M.H., Noorollahi Y., 2017, GA/AHP-based optimal design of a hybrid CCHP system 

considering economy, energy and emission. Energy and Buildings, 138, 309–317. 

Yang G., Zhu J., Yang C., Liu Y., Qiu Y., 2023, Flexible Optimal Scheduling of Electric Power for Hydrogen-

Based Methanol Synthesis System Adapting to Fluctuating Wind Power. Electric Power Construction, 44, 

149–162. (in Chinese) 

Zheng Y., You S., Li X., Bindner H.W., Münster M., 2022, Data-driven robust optimization for optimal scheduling 

of power to methanol. Energy Conversion and Management, 256, 115338. 

516


	0252.pdf
	LSTM-Driven Predictive Scheduling for Green Energy-Hydrogen-Methanol Integrated System